Pokemon games have emerged as a surprising but effective benchmark for evaluating artificial intelligence capabilities, with major AI models from companies like Anthropic and Google now competing to master the 1996 classic. These nostalgic Game Boy adventures provide an ideal testing ground for assessing AI problem-solving abilities, requiring models to maintain focus through complex, open-ended gameplay with ambiguous objectives. The competitions between different AI systems playing through Pokemon Red and Blue have attracted dedicated audiences on Twitch and become significant enough that companies now highlight Pokemon progress when announcing new AI models.
The big picture: Major AI models are playing Pokemon Red and Blue on Twitch livestreams, turning the classic games into an unexpected benchmark for testing AI problem-solving capabilities.
- Anthropic’s Claude 3.7 Sonnet and Google’s Gemini 2.5 Pro Experimental are both attempting to complete the games, with hundreds of viewers watching their progress.
- Anthropic even highlighted its newer model’s ability to progress further in Pokemon compared to its predecessor when announcing Claude 3.7 Sonnet.
Why Pokemon works as an AI benchmark: The Game Boy classics provide an ideal environment for testing AI’s ability to navigate complex, open-ended goals with ambiguity.
- Pokemon’s checkpoint system offers clear progress markers (gym badges) that help showcase each model’s problem-solving abilities.
- Anthropic notes that “the model’s ability to maintain focus and accomplish open-ended goals will help developers build a wide range of state-of-the-art AI agents.”
Current progress: The two AI models have reached different points in their Pokemon journeys, with Gemini currently outpacing Claude.
- Claude 3.7 Sonnet is currently at Mt. Moon after a recent reset, with its best result so far being earning the 3rd badge in Vermillion City.
- Gemini 2.5 Pro Experimental has already earned all eight badges in Pokemon Blue and is now navigating Victory Road, the final area before the Elite Four.
Reading between the lines: While Gemini appears to be performing better, direct comparisons between the models aren’t entirely fair.
- An AI expert on Lesswrong explains that each Twitch stream operates under different conditions, including differences in setup and developer interaction.
- The competitions are drawing significant viewer interest, with 70+ people watching Claude and 100+ watching Gemini at the time of the article’s publication.
I’m a massive Pokémon fan, and now I’m obsessed with AI models like Gemini and Claude trying to complete Pokémon Red and Blue